**[Alibaba Upgrades Real-Time Speech Model Fun-ASR-Realtime: First-Word Latency Reduced to 100ms, Wenzhou Dialect Recognition Accuracy Exceeds 82%]** July 6 news: Alibaba DAMO Academy has reduced the first-word latency of its streaming speech recognition model, Fun-ASR-Realtime, to the millisecond level, achieving near-instant transcription as words are spoken, with accuracy approaching that of offline models. During the recently concluded 100-hour deserted island livestream by Film Hurricane, the model provided real-time subtitle support throughout, even in challenging environments such as outdoor heavy rain and frequent speaker switches. It cumulatively recognized over 60,000 entries, totaling 1.32 million characters. To address the common issue of misinterpretation in real-time speech recognition, the new model enhances contextual awareness, allowing it to dynamically correct errors based on historical dialogue and real-time keywords (e.g., automatically correcting "Ye Lu" to "Night Heron" based on subsequent context). The model currently supports 30 languages and 16 dialects, achieving an average character accuracy rate of 88.62% in dialect tests, with Shanghainese reaching 92.41% and the notoriously difficult Wenzhou dialect achieving 82.74%. The offline version, Fun-ASR-Flash, has claimed the top spot on the word error rate leaderboard of the global AI evaluation platform Artificial Analysis. API services for both models are now available on Alibaba Cloud's ModelScope platform, while the underlying open-source toolkits and models can be accessed on the ModelScope community and GitHub.